On Wed, 15 Feb 2012 14:50:02 +0530, "Srivatsa S. Bhat" <srivatsa.bhat@xxxxxxxxxxxxxxxxxx> wrote:
> > - cpu_clear(smp_processor_id(), mask);
> > - for_each_cpu_mask(cpu, mask)
> > - if (cpu_context(cpu, mm))
> > + for_each_online_cpu(cpu) {
> > + if (cpu != smp_processor_id() && cpu_context(cpu, mm))
> > cpu_context(cpu, mm) = 0;
> > + }
>
>
> Strictly speaking, this one is not a mere cleanup. It causes a subtle change in
> behaviour: earlier, it used to iterate over a local copy of cpu_online_mask, which
> wouldn't change. However, with this patch, it will iterate directly over
> cpu_online_mask, which can change underneath. (The preempt_disable() won't stop
> new CPUs from coming in.. it only prevents CPUs from going offline, that too
> provided that we use stop_machine stuff for CPU offline, which we do currently.)
There's a preempt_disable() around this whole function, so online_mask
can't change.
Same with the others.
> > + mask = *cpu_online_mask;
> > + cpumask_clear(&mask, cpu);
>
>
> This should be cpumask_clear_cpu(cpu, &mask);
Good catch. I copied the bitmask ops, and continually regret it.
I've rolled all these together with your fixes, added your ia64 patch,
and am rebasing to -next now, so I can hand this all across to akpm.
Thanks,
Rusty.
